1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/backing-dev.h>
6 #include <linux/pagemap.h>
7 #include <linux/writeback.h> /* generic_writepages */
8 #include <linux/slab.h>
9 #include <linux/pagevec.h>
10 #include <linux/task_io_accounting_ops.h>
13 #include "mds_client.h"
14 #include <linux/ceph/osd_client.h>
17 * Ceph address space ops.
19 * There are a few funny things going on here.
21 * The page->private field is used to reference a struct
22 * ceph_snap_context for _every_ dirty page. This indicates which
23 * snapshot the page was logically dirtied in, and thus which snap
24 * context needs to be associated with the osd write during writeback.
26 * Similarly, struct ceph_inode_info maintains a set of counters to
27 * count dirty pages on the inode. In the absence of snapshots,
28 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
30 * When a snapshot is taken (that is, when the client receives
31 * notification that a snapshot was taken), each inode with caps and
32 * with dirty pages (dirty pages implies there is a cap) gets a new
33 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
34 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
35 * moved to capsnap->dirty. (Unless a sync write is currently in
36 * progress. In that case, the capsnap is said to be "pending", new
37 * writes cannot start, and the capsnap isn't "finalized" until the
38 * write completes (or fails) and a final size/mtime for the inode for
39 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
41 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
42 * we look for the first capsnap in i_cap_snaps and write out pages in
43 * that snap context _only_. Then we move on to the next capsnap,
44 * eventually reaching the "live" or "head" context (i.e., pages that
45 * are not yet snapped) and are writing the most recently dirtied
48 * Invalidate and so forth must take care to ensure the dirty page
49 * accounting is preserved.
52 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
53 #define CONGESTION_OFF_THRESH(congestion_kb) \
54 (CONGESTION_ON_THRESH(congestion_kb) - \
55 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
57 static inline struct ceph_snap_context *page_snap_context(struct page *page)
59 if (PagePrivate(page))
60 return (void *)page->private;
65 * Dirty a page. Optimistically adjust accounting, on the assumption
66 * that we won't race with invalidate. If we do, readjust.
68 static int ceph_set_page_dirty(struct page *page)
70 struct address_space *mapping = page->mapping;
72 struct ceph_inode_info *ci;
74 struct ceph_snap_context *snapc;
76 if (unlikely(!mapping))
77 return !TestSetPageDirty(page);
79 if (TestSetPageDirty(page)) {
80 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
81 mapping->host, page, page->index);
85 inode = mapping->host;
86 ci = ceph_inode(inode);
89 * Note that we're grabbing a snapc ref here without holding
92 snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
95 spin_lock(&ci->i_ceph_lock);
96 if (ci->i_head_snapc == NULL)
97 ci->i_head_snapc = ceph_get_snap_context(snapc);
98 ++ci->i_wrbuffer_ref_head;
99 if (ci->i_wrbuffer_ref == 0)
101 ++ci->i_wrbuffer_ref;
102 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
103 "snapc %p seq %lld (%d snaps)\n",
104 mapping->host, page, page->index,
105 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
106 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
107 snapc, snapc->seq, snapc->num_snaps);
108 spin_unlock(&ci->i_ceph_lock);
110 /* now adjust page */
111 spin_lock_irq(&mapping->tree_lock);
112 if (page->mapping) { /* Race with truncate? */
113 WARN_ON_ONCE(!PageUptodate(page));
114 account_page_dirtied(page, page->mapping);
115 radix_tree_tag_set(&mapping->page_tree,
116 page_index(page), PAGECACHE_TAG_DIRTY);
119 * Reference snap context in page->private. Also set
120 * PagePrivate so that we get invalidatepage callback.
122 page->private = (unsigned long)snapc;
123 SetPagePrivate(page);
125 dout("ANON set_page_dirty %p (raced truncate?)\n", page);
129 spin_unlock_irq(&mapping->tree_lock);
132 /* whoops, we failed to dirty the page */
133 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
135 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
137 BUG_ON(!PageDirty(page));
142 * If we are truncating the full page (i.e. offset == 0), adjust the
143 * dirty page counters appropriately. Only called if there is private
146 static void ceph_invalidatepage(struct page *page, unsigned long offset)
149 struct ceph_inode_info *ci;
150 struct ceph_snap_context *snapc = page_snap_context(page);
152 BUG_ON(!PageLocked(page));
153 BUG_ON(!PagePrivate(page));
154 BUG_ON(!page->mapping);
156 inode = page->mapping->host;
159 * We can get non-dirty pages here due to races between
160 * set_page_dirty and truncate_complete_page; just spit out a
161 * warning, in case we end up with accounting problems later.
163 if (!PageDirty(page))
164 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
167 ClearPageChecked(page);
169 ci = ceph_inode(inode);
171 dout("%p invalidatepage %p idx %lu full dirty page %lu\n",
172 inode, page, page->index, offset);
173 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
174 ceph_put_snap_context(snapc);
176 ClearPagePrivate(page);
178 dout("%p invalidatepage %p idx %lu partial dirty page\n",
179 inode, page, page->index);
183 /* just a sanity check */
184 static int ceph_releasepage(struct page *page, gfp_t g)
186 struct inode *inode = page->mapping ? page->mapping->host : NULL;
187 dout("%p releasepage %p idx %lu\n", inode, page, page->index);
188 WARN_ON(PageDirty(page));
189 WARN_ON(PagePrivate(page));
194 * read a single page, without unlocking it.
196 static int readpage_nounlock(struct file *filp, struct page *page)
198 struct inode *inode = file_inode(filp);
199 struct ceph_inode_info *ci = ceph_inode(inode);
200 struct ceph_osd_client *osdc =
201 &ceph_inode_to_client(inode)->client->osdc;
203 u64 len = PAGE_CACHE_SIZE;
205 dout("readpage inode %p file %p page %p index %lu\n",
206 inode, filp, page, page->index);
207 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
208 (u64) page_offset(page), &len,
209 ci->i_truncate_seq, ci->i_truncate_size,
217 if (err < PAGE_CACHE_SIZE) {
218 /* zero fill remainder of page */
219 zero_user_segment(page, err, PAGE_CACHE_SIZE);
221 flush_dcache_page(page);
224 SetPageUptodate(page);
227 return err < 0 ? err : 0;
230 static int ceph_readpage(struct file *filp, struct page *page)
232 int r = readpage_nounlock(filp, page);
238 * Finish an async read(ahead) op.
240 static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg)
242 struct inode *inode = req->r_inode;
243 struct ceph_osd_data *osd_data;
244 int rc = req->r_result;
245 int bytes = le32_to_cpu(msg->hdr.data_len);
249 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
251 /* unlock all pages, zeroing any data we didn't read */
252 osd_data = osd_req_op_extent_osd_data(req, 0);
253 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
254 num_pages = calc_pages_for((u64)osd_data->alignment,
255 (u64)osd_data->length);
256 for (i = 0; i < num_pages; i++) {
257 struct page *page = osd_data->pages[i];
259 if (bytes < (int)PAGE_CACHE_SIZE) {
260 /* zero (remainder of) page */
261 int s = bytes < 0 ? 0 : bytes;
262 zero_user_segment(page, s, PAGE_CACHE_SIZE);
264 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
266 flush_dcache_page(page);
267 SetPageUptodate(page);
269 page_cache_release(page);
270 bytes -= PAGE_CACHE_SIZE;
272 kfree(osd_data->pages);
275 static void ceph_unlock_page_vector(struct page **pages, int num_pages)
279 for (i = 0; i < num_pages; i++)
280 unlock_page(pages[i]);
284 * start an async read(ahead) operation. return nr_pages we submitted
285 * a read for on success, or negative error code.
287 static int start_read(struct inode *inode, struct list_head *page_list, int max)
289 struct ceph_osd_client *osdc =
290 &ceph_inode_to_client(inode)->client->osdc;
291 struct ceph_inode_info *ci = ceph_inode(inode);
292 struct page *page = list_entry(page_list->prev, struct page, lru);
293 struct ceph_vino vino;
294 struct ceph_osd_request *req;
303 off = (u64) page_offset(page);
306 next_index = page->index;
307 list_for_each_entry_reverse(page, page_list, lru) {
308 if (page->index != next_index)
312 if (max && nr_pages == max)
315 len = nr_pages << PAGE_CACHE_SHIFT;
316 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
318 vino = ceph_vino(inode);
319 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
321 CEPH_OSD_FLAG_READ, NULL,
322 ci->i_truncate_seq, ci->i_truncate_size,
327 /* build page vector */
328 nr_pages = calc_pages_for(0, len);
329 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS);
333 for (i = 0; i < nr_pages; ++i) {
334 page = list_entry(page_list->prev, struct page, lru);
335 BUG_ON(PageLocked(page));
336 list_del(&page->lru);
338 dout("start_read %p adding %p idx %lu\n", inode, page,
340 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
342 page_cache_release(page);
343 dout("start_read %p add_to_page_cache failed %p\n",
350 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
351 req->r_callback = finish_read;
352 req->r_inode = inode;
354 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
356 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
357 ret = ceph_osdc_start_request(osdc, req, false);
360 ceph_osdc_put_request(req);
364 ceph_unlock_page_vector(pages, nr_pages);
365 ceph_release_page_vector(pages, nr_pages);
367 ceph_osdc_put_request(req);
373 * Read multiple pages. Leave pages we don't read + unlock in page_list;
374 * the caller (VM) cleans them up.
376 static int ceph_readpages(struct file *file, struct address_space *mapping,
377 struct list_head *page_list, unsigned nr_pages)
379 struct inode *inode = file_inode(file);
380 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
384 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
385 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
388 dout("readpages %p file %p nr_pages %d max %d\n", inode,
391 while (!list_empty(page_list)) {
392 rc = start_read(inode, page_list, max);
398 dout("readpages %p file %p ret %d\n", inode, file, rc);
403 * Get ref for the oldest snapc for an inode with dirty data... that is, the
404 * only snap context we are allowed to write back.
406 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
409 struct ceph_inode_info *ci = ceph_inode(inode);
410 struct ceph_snap_context *snapc = NULL;
411 struct ceph_cap_snap *capsnap = NULL;
413 spin_lock(&ci->i_ceph_lock);
414 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
415 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
416 capsnap->context, capsnap->dirty_pages);
417 if (capsnap->dirty_pages) {
418 snapc = ceph_get_snap_context(capsnap->context);
420 *snap_size = capsnap->size;
424 if (!snapc && ci->i_wrbuffer_ref_head) {
425 snapc = ceph_get_snap_context(ci->i_head_snapc);
426 dout(" head snapc %p has %d dirty pages\n",
427 snapc, ci->i_wrbuffer_ref_head);
429 spin_unlock(&ci->i_ceph_lock);
434 * Write a single page, but leave the page locked.
436 * If we get a write error, set the page error bit, but still adjust the
437 * dirty page accounting (i.e., page is no longer dirty).
439 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
442 struct ceph_inode_info *ci;
443 struct ceph_fs_client *fsc;
444 struct ceph_osd_client *osdc;
445 loff_t page_off = page_offset(page);
446 int len = PAGE_CACHE_SIZE;
449 struct ceph_snap_context *snapc, *oldest;
453 dout("writepage %p idx %lu\n", page, page->index);
455 if (!page->mapping || !page->mapping->host) {
456 dout("writepage %p - no mapping\n", page);
459 inode = page->mapping->host;
460 ci = ceph_inode(inode);
461 fsc = ceph_inode_to_client(inode);
462 osdc = &fsc->client->osdc;
464 /* verify this is a writeable snap context */
465 snapc = page_snap_context(page);
467 dout("writepage %p page %p not dirty?\n", inode, page);
470 oldest = get_oldest_context(inode, &snap_size);
471 if (snapc->seq > oldest->seq) {
472 dout("writepage %p page %p snapc %p not writeable - noop\n",
474 /* we should only noop if called by kswapd */
475 WARN_ON((current->flags & PF_MEMALLOC) == 0);
476 ceph_put_snap_context(oldest);
479 ceph_put_snap_context(oldest);
481 /* is this a partial page at end of file? */
485 i_size = i_size_read(inode);
486 if (i_size < page_off + len)
487 len = i_size - page_off;
489 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
490 inode, page, page->index, page_off, len, snapc);
492 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
494 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
495 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
497 set_page_writeback(page);
498 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
499 &ci->i_layout, snapc,
501 ci->i_truncate_seq, ci->i_truncate_size,
502 &inode->i_mtime, &page, 1);
504 dout("writepage setting page/mapping error %d %p\n", err, page);
506 mapping_set_error(&inode->i_data, err);
508 wbc->pages_skipped++;
510 dout("writepage cleaned page %p\n", page);
511 err = 0; /* vfs expects us to return 0 */
514 ClearPagePrivate(page);
515 end_page_writeback(page);
516 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
517 ceph_put_snap_context(snapc); /* page's reference */
522 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
525 struct inode *inode = page->mapping->host;
528 err = writepage_nounlock(page, wbc);
536 * lame release_pages helper. release_pages() isn't exported to
539 static void ceph_release_pages(struct page **pages, int num)
544 pagevec_init(&pvec, 0);
545 for (i = 0; i < num; i++) {
546 if (pagevec_add(&pvec, pages[i]) == 0)
547 pagevec_release(&pvec);
549 pagevec_release(&pvec);
554 * async writeback completion handler.
556 * If we get an error, set the mapping error bit, but not the individual
559 static void writepages_finish(struct ceph_osd_request *req,
560 struct ceph_msg *msg)
562 struct inode *inode = req->r_inode;
563 struct ceph_inode_info *ci = ceph_inode(inode);
564 struct ceph_osd_data *osd_data;
569 struct ceph_snap_context *snapc = req->r_snapc;
570 struct address_space *mapping = inode->i_mapping;
571 int rc = req->r_result;
572 u64 bytes = req->r_ops[0].extent.length;
573 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
575 unsigned issued = ceph_caps_issued(ci);
577 osd_data = osd_req_op_extent_osd_data(req, 0);
578 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
579 num_pages = calc_pages_for((u64)osd_data->alignment,
580 (u64)osd_data->length);
583 * Assume we wrote the pages we originally sent. The
584 * osd might reply with fewer pages if our writeback
585 * raced with a truncation and was adjusted at the osd,
586 * so don't believe the reply.
591 mapping_set_error(mapping, rc);
593 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
594 inode, rc, bytes, wrote);
596 /* clean all pages */
597 for (i = 0; i < num_pages; i++) {
598 page = osd_data->pages[i];
600 WARN_ON(!PageUptodate(page));
603 atomic_long_dec_return(&fsc->writeback_count);
605 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
606 clear_bdi_congested(&fsc->backing_dev_info,
609 ceph_put_snap_context(page_snap_context(page));
611 ClearPagePrivate(page);
612 dout("unlocking %d %p\n", i, page);
613 end_page_writeback(page);
616 * We lost the cache cap, need to truncate the page before
617 * it is unlocked, otherwise we'd truncate it later in the
618 * page truncation thread, possibly losing some data that
621 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
622 generic_error_remove_page(inode->i_mapping, page);
626 dout("%p wrote+cleaned %d pages\n", inode, wrote);
627 ceph_put_wrbuffer_cap_refs(ci, num_pages, snapc);
629 ceph_release_pages(osd_data->pages, num_pages);
630 if (osd_data->pages_from_pool)
631 mempool_free(osd_data->pages,
632 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
634 kfree(osd_data->pages);
635 ceph_osdc_put_request(req);
638 static struct ceph_osd_request *
639 ceph_writepages_osd_request(struct inode *inode, u64 offset, u64 *len,
640 struct ceph_snap_context *snapc, int num_ops)
642 struct ceph_fs_client *fsc;
643 struct ceph_inode_info *ci;
644 struct ceph_vino vino;
646 fsc = ceph_inode_to_client(inode);
647 ci = ceph_inode(inode);
648 vino = ceph_vino(inode);
649 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
651 return ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
652 vino, offset, len, num_ops, CEPH_OSD_OP_WRITE,
653 CEPH_OSD_FLAG_WRITE|CEPH_OSD_FLAG_ONDISK,
654 snapc, ci->i_truncate_seq, ci->i_truncate_size, true);
658 * initiate async writeback
660 static int ceph_writepages_start(struct address_space *mapping,
661 struct writeback_control *wbc)
663 struct inode *inode = mapping->host;
664 struct ceph_inode_info *ci = ceph_inode(inode);
665 struct ceph_fs_client *fsc;
666 pgoff_t index, start, end;
669 pgoff_t max_pages = 0, max_pages_ever = 0;
670 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
674 unsigned wsize = 1 << inode->i_blkbits;
675 struct ceph_osd_request *req = NULL;
680 * Include a 'sync' in the OSD request if this is a data
681 * integrity write (e.g., O_SYNC write or fsync()), or if our
682 * cap is being revoked.
684 do_sync = wbc->sync_mode == WB_SYNC_ALL;
685 if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
687 dout("writepages_start %p dosync=%d (mode=%s)\n",
689 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
690 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
692 fsc = ceph_inode_to_client(inode);
693 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
694 pr_warning("writepage_start %p on forced umount\n", inode);
695 return -EIO; /* we're in a forced umount, don't write! */
697 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
698 wsize = fsc->mount_options->wsize;
699 if (wsize < PAGE_CACHE_SIZE)
700 wsize = PAGE_CACHE_SIZE;
701 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
703 pagevec_init(&pvec, 0);
705 /* where to start/end? */
706 if (wbc->range_cyclic) {
707 start = mapping->writeback_index; /* Start from prev offset */
709 dout(" cyclic, start at %lu\n", start);
711 start = wbc->range_start >> PAGE_CACHE_SHIFT;
712 end = wbc->range_end >> PAGE_CACHE_SHIFT;
713 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
716 dout(" not cyclic, %lu to %lu\n", start, end);
721 /* find oldest snap context with dirty data */
722 ceph_put_snap_context(snapc);
724 snapc = get_oldest_context(inode, &snap_size);
726 /* hmm, why does writepages get called when there
728 dout(" no snap context with dirty data?\n");
732 snap_size = i_size_read(inode);
733 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
734 snapc, snapc->seq, snapc->num_snaps);
735 if (last_snapc && snapc != last_snapc) {
736 /* if we switched to a newer snapc, restart our scan at the
737 * start of the original file range. */
738 dout(" snapc differs from last pass, restarting at %lu\n",
744 while (!done && index <= end) {
745 int num_ops = do_sync ? 2 : 1;
746 struct ceph_vino vino;
750 int pvec_pages, locked_pages;
751 struct page **pages = NULL;
752 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
760 max_pages = max_pages_ever;
764 want = min(end - index,
765 min((pgoff_t)PAGEVEC_SIZE,
766 max_pages - (pgoff_t)locked_pages) - 1)
768 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
771 dout("pagevec_lookup_tag got %d\n", pvec_pages);
772 if (!pvec_pages && !locked_pages)
774 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
775 page = pvec.pages[i];
776 dout("? %p idx %lu\n", page, page->index);
777 if (locked_pages == 0)
778 lock_page(page); /* first page */
779 else if (!trylock_page(page))
782 /* only dirty pages, or our accounting breaks */
783 if (unlikely(!PageDirty(page)) ||
784 unlikely(page->mapping != mapping)) {
785 dout("!dirty or !mapping %p\n", page);
789 if (!wbc->range_cyclic && page->index > end) {
790 dout("end of range %p\n", page);
795 if (next && (page->index != next)) {
796 dout("not consecutive %p\n", page);
800 if (wbc->sync_mode != WB_SYNC_NONE) {
801 dout("waiting on writeback %p\n", page);
802 wait_on_page_writeback(page);
804 if (page_offset(page) >= snap_size) {
805 dout("%p page eof %llu\n", page, snap_size);
810 if (PageWriteback(page)) {
811 dout("%p under writeback\n", page);
816 /* only if matching snap context */
817 pgsnapc = page_snap_context(page);
818 if (pgsnapc->seq > snapc->seq) {
819 dout("page snapc %p %lld > oldest %p %lld\n",
820 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
823 continue; /* keep looking for snap */
827 if (!clear_page_dirty_for_io(page)) {
828 dout("%p !clear_page_dirty_for_io\n", page);
834 * We have something to write. If this is
835 * the first locked page this time through,
836 * allocate an osd request and a page array
839 if (locked_pages == 0) {
844 /* prepare async write request */
845 offset = (u64)page_offset(page);
847 req = ceph_writepages_osd_request(inode,
857 req->r_callback = writepages_finish;
858 req->r_inode = inode;
860 max_pages = calc_pages_for(0, (u64)len);
861 size = max_pages * sizeof (*pages);
862 pages = kmalloc(size, GFP_NOFS);
864 pool = fsc->wb_pagevec_pool;
865 pages = mempool_alloc(pool, GFP_NOFS);
870 /* note position of first page in pvec */
873 dout("%p will write page %p idx %lu\n",
874 inode, page, page->index);
877 atomic_long_inc_return(&fsc->writeback_count);
878 if (writeback_stat > CONGESTION_ON_THRESH(
879 fsc->mount_options->congestion_kb)) {
880 set_bdi_congested(&fsc->backing_dev_info,
884 set_page_writeback(page);
885 pages[locked_pages] = page;
887 next = page->index + 1;
890 /* did we get anything? */
892 goto release_pvec_pages;
895 BUG_ON(!locked_pages || first < 0);
897 if (pvec_pages && i == pvec_pages &&
898 locked_pages < max_pages) {
899 dout("reached end pvec, trying for more\n");
900 pagevec_reinit(&pvec);
904 /* shift unused pages over in the pvec... we
905 * will need to release them below. */
906 for (j = i; j < pvec_pages; j++) {
907 dout(" pvec leftover page %p\n",
909 pvec.pages[j-i+first] = pvec.pages[j];
914 /* Format the osd request message and submit the write */
916 offset = page_offset(pages[0]);
917 len = min(snap_size - offset,
918 (u64)locked_pages << PAGE_CACHE_SHIFT);
919 dout("writepages got %d pages at %llu~%llu\n",
920 locked_pages, offset, len);
922 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
925 pages = NULL; /* request message now owns the pages array */
928 /* Update the write op length in case we changed it */
930 osd_req_op_extent_update(req, 0, len);
932 vino = ceph_vino(inode);
933 ceph_osdc_build_request(req, offset, snapc, vino.snap,
936 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
942 wbc->nr_to_write -= locked_pages;
943 if (wbc->nr_to_write <= 0)
947 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
948 pvec.nr ? pvec.pages[0] : NULL);
949 pagevec_release(&pvec);
951 if (locked_pages && !done)
955 if (should_loop && !done) {
956 /* more to do; loop back to beginning of file */
957 dout("writepages looping back to beginning of file\n");
963 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
964 mapping->writeback_index = index;
968 ceph_osdc_put_request(req);
969 ceph_put_snap_context(snapc);
970 dout("writepages done, rc = %d\n", rc);
977 * See if a given @snapc is either writeable, or already written.
979 static int context_is_writeable_or_written(struct inode *inode,
980 struct ceph_snap_context *snapc)
982 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
983 int ret = !oldest || snapc->seq <= oldest->seq;
985 ceph_put_snap_context(oldest);
990 * We are only allowed to write into/dirty the page if the page is
991 * clean, or already dirty within the same snap context.
993 * called with page locked.
994 * return success with page locked,
995 * or any failure (incl -EAGAIN) with page unlocked.
997 static int ceph_update_writeable_page(struct file *file,
998 loff_t pos, unsigned len,
1001 struct inode *inode = file_inode(file);
1002 struct ceph_inode_info *ci = ceph_inode(inode);
1003 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1004 loff_t page_off = pos & PAGE_CACHE_MASK;
1005 int pos_in_page = pos & ~PAGE_CACHE_MASK;
1006 int end_in_page = pos_in_page + len;
1009 struct ceph_snap_context *snapc, *oldest;
1012 /* writepages currently holds page lock, but if we change that later, */
1013 wait_on_page_writeback(page);
1015 /* check snap context */
1016 BUG_ON(!ci->i_snap_realm);
1017 down_read(&mdsc->snap_rwsem);
1018 BUG_ON(!ci->i_snap_realm->cached_context);
1019 snapc = page_snap_context(page);
1020 if (snapc && snapc != ci->i_head_snapc) {
1022 * this page is already dirty in another (older) snap
1023 * context! is it writeable now?
1025 oldest = get_oldest_context(inode, NULL);
1026 up_read(&mdsc->snap_rwsem);
1028 if (snapc->seq > oldest->seq) {
1029 ceph_put_snap_context(oldest);
1030 dout(" page %p snapc %p not current or oldest\n",
1033 * queue for writeback, and wait for snapc to
1034 * be writeable or written
1036 snapc = ceph_get_snap_context(snapc);
1038 ceph_queue_writeback(inode);
1039 r = wait_event_interruptible(ci->i_cap_wq,
1040 context_is_writeable_or_written(inode, snapc));
1041 ceph_put_snap_context(snapc);
1042 if (r == -ERESTARTSYS)
1046 ceph_put_snap_context(oldest);
1048 /* yay, writeable, do it now (without dropping page lock) */
1049 dout(" page %p snapc %p not current, but oldest\n",
1051 if (!clear_page_dirty_for_io(page))
1053 r = writepage_nounlock(page, NULL);
1059 if (PageUptodate(page)) {
1060 dout(" page %p already uptodate\n", page);
1065 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1068 /* past end of file? */
1069 i_size = inode->i_size; /* caller holds i_mutex */
1071 if (i_size + len > inode->i_sb->s_maxbytes) {
1072 /* file is too big */
1077 if (page_off >= i_size ||
1078 (pos_in_page == 0 && (pos+len) >= i_size &&
1079 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1080 dout(" zeroing %p 0 - %d and %d - %d\n",
1081 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1082 zero_user_segments(page,
1084 end_in_page, PAGE_CACHE_SIZE);
1088 /* we need to read it. */
1089 up_read(&mdsc->snap_rwsem);
1090 r = readpage_nounlock(file, page);
1096 up_read(&mdsc->snap_rwsem);
1103 * We are only allowed to write into/dirty the page if the page is
1104 * clean, or already dirty within the same snap context.
1106 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1107 loff_t pos, unsigned len, unsigned flags,
1108 struct page **pagep, void **fsdata)
1110 struct inode *inode = file_inode(file);
1112 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1117 page = grab_cache_page_write_begin(mapping, index, 0);
1122 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1123 inode, page, (int)pos, (int)len);
1125 r = ceph_update_writeable_page(file, pos, len, page);
1126 } while (r == -EAGAIN);
1132 * we don't do anything in here that simple_write_end doesn't do
1133 * except adjust dirty page accounting and drop read lock on
1136 static int ceph_write_end(struct file *file, struct address_space *mapping,
1137 loff_t pos, unsigned len, unsigned copied,
1138 struct page *page, void *fsdata)
1140 struct inode *inode = file_inode(file);
1141 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1142 struct ceph_mds_client *mdsc = fsc->mdsc;
1143 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1146 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1147 inode, page, (int)pos, (int)copied, (int)len);
1149 /* zero the stale part of the page if we did a short copy */
1151 zero_user_segment(page, from+copied, len);
1153 /* did file size increase? */
1154 /* (no need for i_size_read(); we caller holds i_mutex */
1155 if (pos+copied > inode->i_size)
1156 check_cap = ceph_inode_set_size(inode, pos+copied);
1158 if (!PageUptodate(page))
1159 SetPageUptodate(page);
1161 set_page_dirty(page);
1164 up_read(&mdsc->snap_rwsem);
1165 page_cache_release(page);
1168 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1174 * we set .direct_IO to indicate direct io is supported, but since we
1175 * intercept O_DIRECT reads and writes early, this function should
1178 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1179 const struct iovec *iov,
1180 loff_t pos, unsigned long nr_segs)
1186 const struct address_space_operations ceph_aops = {
1187 .readpage = ceph_readpage,
1188 .readpages = ceph_readpages,
1189 .writepage = ceph_writepage,
1190 .writepages = ceph_writepages_start,
1191 .write_begin = ceph_write_begin,
1192 .write_end = ceph_write_end,
1193 .set_page_dirty = ceph_set_page_dirty,
1194 .invalidatepage = ceph_invalidatepage,
1195 .releasepage = ceph_releasepage,
1196 .direct_IO = ceph_direct_io,
1205 * Reuse write_begin here for simplicity.
1207 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1209 struct inode *inode = file_inode(vma->vm_file);
1210 struct page *page = vmf->page;
1211 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1212 loff_t off = page_offset(page);
1216 /* Update time before taking page lock */
1217 file_update_time(vma->vm_file);
1219 size = i_size_read(inode);
1220 if (off + PAGE_CACHE_SIZE <= size)
1221 len = PAGE_CACHE_SIZE;
1223 len = size & ~PAGE_CACHE_MASK;
1225 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1226 off, len, page, page->index);
1230 ret = VM_FAULT_NOPAGE;
1232 (page->mapping != inode->i_mapping))
1235 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1237 /* success. we'll keep the page locked. */
1238 set_page_dirty(page);
1239 up_read(&mdsc->snap_rwsem);
1240 ret = VM_FAULT_LOCKED;
1245 ret = VM_FAULT_SIGBUS;
1248 dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1249 if (ret != VM_FAULT_LOCKED)
1254 static struct vm_operations_struct ceph_vmops = {
1255 .fault = filemap_fault,
1256 .page_mkwrite = ceph_page_mkwrite,
1257 .remap_pages = generic_file_remap_pages,
1260 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1262 struct address_space *mapping = file->f_mapping;
1264 if (!mapping->a_ops->readpage)
1266 file_accessed(file);
1267 vma->vm_ops = &ceph_vmops;